[Thread reset] GCC4 vs binutils

[Thread reset] GCC4 vs binutils

[Dave Korn]

[ Time to start a fresh thread, since something's come up. ]

  In the thread at

        http://cygwin.com/ml/cygwin/2009-02/threads.html#00448

Chuck reported a problem with cygmagic-1.dll causing a segfault when rebased.
 To cut to the end of the thread, this turns out to be owing to a bug that was
fixed upstream in ld, unfortunately just a few weeks after the current
binutils was released:

> http://sourceware.org/ml/binutils/2008-07/msg00372.html

  This problem affects potentially all maintainers and end-users who compile
DLLs using -static-libgcc.  Otherwise, it would have been the case that only
users who wanted to build the compiler themselves from sources would need to
worry about having a more recent binutils, but this problem can bite anybody,
so I guess we need to get a newer binutils out there.  Aaaaand, as long as
we're making plans for silencing that ld warning, we should try and get that
change in there first.

  CGF, how are you fixed for time right now?  Will you be able to do a release
sometime in the next week say?  But not so quickly that we don't have time to
get a quick patch in upstream to shut up the auto-import warning?  On a side
issue, are you going to want to handle the MinGW cross-binutils we'll be
needing for the cross-compiler further on down the line?

    cheers,
      DaveK

Re: [Thread reset] GCC4 vs binutils

[Charles Wilson]

.Dave Korn wrote:
 Aaaaand, as long as
> we're making plans for silencing that ld warning, we should try and get that
> change in there first.

Background:

The current behavior wrt auto-import (enabled by default, but complain
-- info, not warn -- if used without explicit --enable-auto-import)
happened here:
http://sourceware.org/ml/binutils/2002-05/msg00703.html
and extends to ALL pe platforms.

Prior to that, auto-import was disabled by default, and if enabled,
issued a warning if used. This dates back to when auto-import was first
added to ld:
http://sourceware.org/ml/binutils/2001-08/msg00040.html

So, in short: auto-import has been available for 7.5 years, and has been
the default (but whines with "info" message) on all pe platforms for
just under 7 years.

I don't think anyone will complain about no longer whining. It's an
established tool in the gcc/ld-on-pe toolbox.

Dave, do you want to propose the patch on binutils, or shall I?

--
Chuck

Re: [Thread reset] GCC4 vs binutils

[Dave Korn]

Charles Wilson wrote:
> .Dave Korn wrote:
>  Aaaaand, as long as
>> we're making plans for silencing that ld warning, we should try and get that
>> change in there first.
> 
> Background:
> 
> The current behavior wrt auto-import (enabled by default, but complain
> -- info, not warn -- if used without explicit --enable-auto-import)
> happened here:
> http://sourceware.org/ml/binutils/2002-05/msg00703.html
> and extends to ALL pe platforms.

  See also http://sourceware.org/bugzilla/show_bug.cgi?id=4844.

> Dave, do you want to propose the patch on binutils, or shall I?

  I've already started testing one :)

    cheers,
      DaveK

Re: [Thread reset] GCC4 vs binutils

[Dave Korn]

Charles Wilson wrote:

> So, in short: auto-import has been available for 7.5 years, and has been
> the default (but whines with "info" message) on all pe platforms for
> just under 7 years.
> 
> I don't think anyone will complain about no longer whining. It's an
> established tool in the gcc/ld-on-pe toolbox.

  Just to restate FTA what was posted elsewhere in another thread: it is also
a minor change in behaviour WRT allocation of sections.

> Dave, do you want to propose the patch on binutils, or shall I?

  Now posted at http://sourceware.org/ml/binutils/2009-02/msg00341.html.

    cheers,
      DaveK